I just realized that
#vibecoding is just the software-development equivalent of cooking at
#McDonalds.
The output of
#LLM is
#junkcode.
While
#programming,
#hackers retain full control of the means of production, firmly anchored on their necks.
In the last forty years, this turned programming into an wealthy carrier that attracted greedy people because the only way corporation had to obliviate awareness of such unprecedented political leverage, was to pay developers relarively high salaries while they were building the infrastructure of their own oppression
With
#vibecoding, means of production go back into capitals hands: novice produce nicely looking software without acquiring any valuable skill, and senior developers leverage their (hardly acquired) experience to "drive the tool like a younger intern", alienating¹ themselves, loosing their skills while providing further "training data"² to the capital owners
Note how I'm not strictly talking about employers: if you work for a company that push a
#CodingAssistant from a third party (usually a
#BigTech from the
#USA), your company is doomed too, as they are giving their most valuable asset (your skills and the business experience encoded in their source code) away.
Yet the point is that junk code is to society what junk food is to public health: a burden that mostly affect the poor, not the rich.
Indeed rich people can pay for fine restaurants and healthy food, while the poorest are forced to eat the cheapest slop they can afford, further enriching the companies that sell it and pay low wages to their employees.
In the same way, the users' of vibecoded software will be those who can't afford high quality software. And vibecoders will be those who can't afford to learn how to code (that requires time and energy, and thus money)
So while
#vibecoding is marketed as "the democratization of programming", such
#propaganda hides the opposite process: if vibecoding keep spreading, programming will become a service to rent under the full control of a handful of companies that will be able to inject any vulnerability or backdoor into the junk code that nobody could actually read.
Paradoxically, those who now resist to the fear of missing out and preserve their skills might gain even higher wages in the future, while those who follow the mob will discover themselves among the replaceble members of the
reserve army of labour, together with McDonald's chefs, forced to ~eat~ depend on junk code.
What about hackers
Hackers will keep programming on their own. Not much for the fun of encoding an insight into a sequence of symbols a compiler can crunch (a process that can be just as frustrating as it's rewarding in term of knowledge) but because they understands both the technicalities of these tools (no
#AI really understand its own output, so it can only work insofar the requirements replicated many existing software, but without any quality or security assurance) and the politics of the corporations that build and operate such tools.
In the long run, the social contract behind
#FreeSoftware will evolve to avoid both contamination from junk code and contribution to the training dataset.
With everything else equal, we could have new
#junkfree stacks, designed to be both human friendly and hostile to corporations: simpler operating system, programming language and protocols.
Unless, obviously, these
#BigTech will somehow manage to outlaw programmable computing device they cannot control, probably in the name of users' security or children protection.
_____
¹
#Cybernetic alienation is the process of reducing human (awareness of)
#autonomy.
#Antropic gaslight the issue framing it as an issue of personal empowerment that can be addressed at design level, but given how people are trained to treat other people as tools and to treat interactive software as people, you see the issue is systemic to
#LLM usage (at least as long as they are programmed to pass the
#Turing test and fool humans about their nature).
² talking about "training data" is alienating by itself, as it project a human experience over an unrelated mechanical process. Instead of "training data" we should talk about "source data", as the models are nothing more than executable expressed as numeric matrices and designed to be executed by specific custom-built architectures that are improperly called "inference engine" (or even worse,
#NeuralNetworks) while they are just statistically programmable vector mapping machines.
karinlustenberger
Comic Crusaders
U3dn
Marko
Giacomo Tesio
Johannes W. Dietrich
µP
Ball Shaped Man
Nicolas Zentz